| Network Working Group | T. C. Player |
| Internet-Draft | Spirent Communications |
| Intended status: Informational | D. Newman |
| Expires: October 20, 2011 | Network Test |
| April 18, 2011 |
Bridge Out: Benchmarking Methodology Extensions for Data Center Bridging Devices
draft-player-dcb-benchmarking-04.txt
Existing benchmarking methodologies are based on the assumption that networking devices will impartially drop network traffic at their performance limits. Data Center Bridging (DCB) devices, however, will attempt to throttle prioritized traffic from network endpoints before those limits are reached in order to minimize the probability of frame loss for high value traffic. Hence, existing methodologies based around indiscriminate frame loss are inappropriate for DCB devices. This document takes the basic benchmarking ideas based on loss and extends them to support "lossless" Ethernet devices.
The IETF has been notified of intellectual property rights claimed in regard to some or all of the specification contained in this document. For more information consult the online list of claimed rights.
This Internet-Draft is submitted in full conformance with the provisions of BCP 78 and BCP 79.
Internet-Drafts are working documents of the Internet Engineering Task Force (IETF). Note that other groups may also distribute working documents as Internet-Drafts. The list of current Internet- Drafts is at http://datatracker.ietf.org/drafts/current/.
Internet-Drafts are draft documents valid for a maximum of six months and may be updated, replaced, or obsoleted by other documents at any time. It is inappropriate to use Internet-Drafts as reference material or to cite them other than as "work in progress."
This Internet-Draft will expire on October 20, 2011.
Copyright (c) 2011 IETF Trust and the persons identified as the document authors. All rights reserved.
This document is subject to BCP 78 and the IETF Trust's Legal Provisions Relating to IETF Documents (http://trustee.ietf.org/license-info) in effect on the date of publication of this document. Please review these documents carefully, as they describe your rights and restrictions with respect to this document. Code Components extracted from this document must include Simplified BSD License text as described in Section 4.e of the Trust Legal Provisions and are provided without warranty as described in the Simplified BSD License.
This document is intended to provide a methodology for benchmarking Data Center Bridging (DCB) devices that support Priority-based Flow Control (PFC). It extends the methodologies already defined in [RFC2544] and [RFC2889].
This memo primarily deals with devices which use Priority-based Flow Control, as defined in IEEE specification 802.1Qbb, to actively manage the transmission rate of multiple classes of traffic in order to minimize forwarding delay and frame loss for high priority traffic.
The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in [RFC2119].
As the terminology used by [RFC4689] is specific to IP layer testing, a number of existing terms require clarification when used in the DCB benchmarking context. Additionally, a number of new terms are also presented to clarify concepts not clearly defined within the scope of [RFC4689].
Data Center Bridging devices SHOULD be tested with multiple Classifications. Testing with a single Classification provides no means to test and measure a device's ability to differentiate forwarding behavior for different traffic classes.
For devices capable of forwarding traffic at line rate, explicit congestion MUST be created via the test tool to benchmark queue performance. Possible methods for accomplishing this on a DUT with n ports include, but are not limited to:
The lock-step traffic pattern, as described in section 5.1.3 of [RFC2889], is specifically NOT required for DCB testing for two reasons:
This testing document does not mandate the use of any particular frame format for testing. Any frame that can be legally forwarded by the DUT/SUT MAY be used provided that the test instrument can make the following distinctions for each frame:
To accurately measure the performance of a Priority-based Flow Control capable DUT, the test tool MUST be able to respond to PFC pause frames. Additionally, the test tool MUST respond to all received pause frames in the time period specified in the IEEE 802.1Qbb specification.
This document extends the general test setup described in section 3 of [RFC2889] and section 6 of [RFC2544] to the benchmarking of Data Center Ethernet switching devices. [RFC2889] and [RFC2544] describe benchmarking methodologies for networking devices that intentionally drop frames at their performance limits. In DCB networks, the DUT will transmit PFC Pause frames as a Congestion Management method to throttle network endpoints, thus minimizing the probability of frame loss in the network.
Since DCB devices are expected to support multiple traffic Classifications, it is RECOMMENDED to benchmark DCB devices with multiple Classification Groups.
The RECOMMENDED trial duration is 300 seconds. However other durations MAY be used. Additionally, a running trial MAY be aborted once the test tool determines that the currently running trial has failed, e.g. QoS bounds exceeded, packet loss detected on a lossless queue, etc.
Packet Conformance MUST be determined for all test frames on a per frame basis. The method specified for measuring Latency in [RFC2544], e.g. measuring the latency of a single test frame in a traffic flow, is unsuitable for DCB benchmarking.
Multiple methods exist for measuring the time it takes a test frame to be forwarded by a DUT. [RFC1242] describes different methods for measuring the latency of "store and forward" and "bit forwarding" devices. Direct comparisons of results obtained using these different methods are inappropriate. Regardless of DCB device architecture, the only RECOMMENDED method for measuring the time it takes a DUT to forward a test frame is the bit forwarding ("cut-through") method as described in [RFC1242].
The recommended frame sizes for Ethernet testing are 64, 128, 256, 512, 1024, 1280, 1518, 4096, 8192, and 9216 as per [RFC5180]. Note that these frame sizes include the Ethernet CRC and VLAN header.
FCoE test traffic introduces a number of frame size constraints that make the default frame sizes specified in [RFC5180] unusable:
Due to these constraints, the recommended frame sizes for FCoE testing are 128, 256, 512, 1024, 1280, 1520, 2176, and the smallest FCoE frame size supported by the test tool. Note that these frame sizes include both the Ethernet CRC and VLAN header.
As per [RFC2285], the burst size specifies the number of test frames in a burst. To simulate bursty traffic, the test tool MAY send a burst of test traffic with the minimum, legal Inter-Frame Gap (IFG) between frames in the burst followed by a larger Inter-Burst Gap (IBG) between sequential bursts. Note that burst sizes are only applicable to test traffic when the Offered Load of the test ports is less than the Maximum Offered Load (MOL) of those ports. Additionally, a burst size of 1 specifies a constant load, e.g. non-bursty traffic.
To determine the amount of time required for the DUT to respond to priority-based flow control pause frames.
The following parameters MUST be defined. Each variable is configured with the following considerations.
The test tool SHOULD generate test traffic for at least 30 seconds before sending any XOFF frame in order for the DUT to reach a steady-state forwarding condition. The test tool then transmits one or more XOFF frames on one or more ports. Each XOFF frame SHOULD instruct the DUT to pause one or more of the Classification Groups currently being forwarded by the DUT. The test tool MAY optionally send a XON frame to instruct the DUT to resume transmission.
The following measurements MUST be reported for each test port and codepoint involved in the test.
TBD
To determine the Queueput for one or more Traffic Classifications of a DUT using priority flow control.
The following parameters MUST be defined. Each variable is configured with the following considerations.
A search algorithm is used to determine the Queueput for each Classification Group. If Queue Congestion is detected for a Classification Group during a trial, then the Intended Vector for the Classification Group MUST be reduced for the subsequent trial. If a Conformance Vector is specified for the test and Non-conformant frames are received during a trial, then the Intended Vector SHOULD be reduced for the subsequent trial. The algorithm MUST adjust the Intended Vector for each Classification Group. The search algorithms for each Classification Group MAY be run in parallel. The test continues until all Classification Groups in the test have converged on a discrete Queueput value.
The Queueput for each Classification MUST be reported in either N-octet frames per second or bits per second.
If a Conformance Vector is specified for a Classification Group, any Non-conformant frames MUST be reported.
The number of PFC pause frames transmitted by the DUT for each code-point in the Codepoint Set MUST be reported for each test port.
The total pause time observed by the tester for each code-point in the Codepoint Set MUST be reported for each test port.
Any frame loss observed for test traffic using PFC enabled codepoints MUST be reported. Any frame loss observed for test traffic using non-PFC enabled codepoints on uncongested egress ports SHOULD be reported, as that indicates the DUT is performing Head of Line Blocking (HOLB).
TBD
To determine the maximum forwarding rate of one or more PFC queues on a PFC capable DUT.
Maximum Forwarding Rate is conceptually similar to the measurement in [RFC2285] but works on a per-Classification basis in a DCB context. The following parameters MUST be defined. Each variable is configured with the following considerations.
The tester should iterate across all configured permutations of frame size, burst size, and Intended Vector for all Classification Groups.
The forwarding rate of each Classification Group MUST be reported as the number of N-octet test frames per second the DUT correctly forwards to the proper egress port.
The maximum forwarding rate for each Classification Group MUST be reported as the highest recorded forwarding rate from the set of all iterations.
Both the Intended and Offered Vector of each Classification Group MUST be reported.
If a Conformance Vector is specified for a Classification Group, any Non-conformant frames MUST be reported.
The number of PFC pause frames transmitted by the DUT for each code-point in the Codepoint Set MUST be reported.
The total pause time observed by the tester for each code-point in the Codepoint Set MUST be reported.
TBD
To determine the delta between the maximum forwarding rate of a DUT and the point where the DUT ceases to use PFC to manage priority queues.
The following parameters MUST be defined. Each variable is configured with the following considerations.
The initial trial SHOULD begin with an Intended Load equal to or greater than the Maximum Forwarding Rate of the DUT/SUT. For each subsequent trial, the aggregate load is reduced until the DUT is observed to complete a trial without activating any Congestion Management methods.
The Intended and Offered Vector for each Classification Group MUST be reported.
If a Conformance Vector is specified for a Classification Group, any Non-conformant frames MUST be reported.
The number of PFC pause frames transmitted by the DUT for each code-point in the Codepoint Set MUST be reported.
The total pause time observed by the tester for each code-point in the Codepoint Set MUST be reported.
Any frame loss observed for test traffic using PFC enabled codepoints MUST be reported. Any frame loss observed for test traffic using non-PFC enabled codepoints on uncongested egress ports SHOULD be reported, as that indicates the DUT is performing Head of Line Blocking (HOLB).
TBD
To determine the maximum duration a DUT can forward test traffic with minimum Inter-Frame Gap on one or more PFC queues without using Congestion Management.
The following parameters MUST be defined. Each variable is configured with the following considerations
A search algorithm is used to determine the maximum duration in seconds for which the configured Classification Profile can be forwarded by the DUT without active Congestion Management. If Congestion Management is detected during an iteration, then the duration MUST be reduced for the next iteration.
The Intended and Offered Vector for each Classification Group MUST be reported.
If a Conformance Vector is specified for a Classification Group, any Non-conformant frames MUST be reported.
The number of PFC pause frames transmitted by the DUT for each code-point in the Codepoint Set MUST be reported.
The total pause time observed by the tester for each code-point in the Codepoint Set MUST be reported.
Any frame loss observed for test traffic using PFC enabled codepoints MUST be reported. Any frame loss observed for test traffic using non-PFC enabled codepoints on uncongested egress ports SHOULD be reported, as that indicates the DUT is performing Head of Line Blocking (HOLB).
TBD
Benchmarking activities as described in this memo are limited to technology characterization using controlled stimuli in a laboratory environment, with dedicated address space and the constraints specified in the sections above.
The benchmarking network topology will be an independent test setup and MUST NOT be connected to devices that may forward the test traffic into a production network, or misroute traffic to the test management network.
Further, benchmarking is performed on a "black-box" basis, relying solely on measurements observable external to the DUT/SUT.
Special capabilities SHOULD NOT exist in the DUT/SUT specifically for benchmarking purposes. Any implications for network security arising from the DUT/SUT SHOULD be identical in the lab and in production networks.
Testers SHOULD use network addresses assigned by IANA for the purpose of testing networks.